Simulated X-ray Projections Research Articles

Pt4: Phantom-4D, an open-source software tool for creating time-evolving 3D phantoms

pt4, an open-source software tool to describe time-evolving phantoms is presented. pt4 allows users to create detailed time-evolving phantoms for testing novel 4D-CT reconstruction algorithms. Ground-truth volumes and simulated X-ray projections can be produced at arbitrary time-points during time-evolution and with customisable pixel dimensions, noise models, and X-ray source trajectories. Phantoms are built up from 3D primitives whose parameters and attenuation can be made arbitrary functions of time. This feature permits both complex continuous and discontinuous time-evolution necessary for thorough testing of 4D-CT reconstruction algorithms. Various phantoms built using pt4 are also presented to demonstrate the versatility of pt4 phantom description. pt4 is written in C++ and is highly parallelised leading to a performant implementation which is feasible to use for up to thousand-of-voxel volume pixel dimensions.

Exact computation of projected sphere centres in cone beam x-ray projections

A method is presented for fitting the projected centres of spheres in cone beam x-ray imaging. By using a suitable coordinate system, the method allows direct and exact calculation of the sphere centre without fitting the projection shape with an ellipse and correcting from the ellipse centre to the sphere centre. Advantages in numerical implementation result from the number of unknown variables being reduced compared to ellipse fits. Additionally, the orientation of the detector relative to the x-ray source can be obtained from fitting the shapes of projections of multiple spheres without knowledge of the positions or dimensions of the spheres. The accuracy of the method is compared to other techniques using simulated x-ray projections.

Fast and accurate pose estimation of additive manufactured objects from few X-ray projections

X-ray Computed Tomography (CT) is a commonly used imaging technique for non-destructive inspection of manufactured objects. However, a full CT scan requires a long acquisition time, making this method unsuitable for inline applications. In contrast to X-ray CT, inspection can be performed directly in the projection space, using simulated X-ray projections of a reference model of the manufactured object. However, to effectively compare simulated and measured projections, an accurate 3D pose estimation of the object and consequent alignment between the measured object and the reference model are crucial. In this paper, we present a fast method to estimate the 3D pose of a measured object based on convolutional neural networks (CNNs). Through experiments on synthetic and measured data, we demonstrate that our method allows estimating the 3D pose of the object with sub-pixel accuracy. Even if very few projections are available, our approach is comparable to CT-based methods for registration, and outperforms state-of-the-art deep learning methods for radiograph-based pose estimation.

Trajectory Optimization in Computed Tomography Based on Object Geometry

We present a new approach for the task of trajectory optimization for X-ray computed tomography (CT) in a robot-based CT (RoboCT) application. An optimized trajectory corresponds to a projection set of predefined size that leads to a 3Dreconstruction with less image-artifacts than other possible sets of the same size. We incorporate prior knowledge through a model of the object and simulated X-ray projections of the interested scenario. We formulate the optimization problem as maximization of an objective function and solve it heuristically by means of a genetic algorithm. As objective function, we use a metric that correlates with the amount of edge information of a projection set, since we assume that this characteristic is important for a reconstruction with low image-artifacts. The metric is based on wavelet analysis of the projections. We carry out computer experiments for a test object made of homogeneous material. The optimized trajectory allows a sharp reconstruction of straight edges and clearly outperforms reconstructions from standard trajectories.

Evaluation of data augmentation via synthetic images for improved breast mass detection on mammograms using deep learning.

We evaluated whether using synthetic mammograms for training data augmentation may reduce the effects of overfitting and increase the performance of a deep learning algorithm for breast mass detection. Synthetic mammograms were generated using in silico procedural analytic breast and breast mass modeling algorithms followed by simulated x-ray projections of the breast models into mammographic images. In silico breast phantoms containing masses were modeled across the four BI-RADS breast density categories, and the masses were modeled with different sizes, shapes, and margins. A Monte Carlo-based x-ray transport simulation code, MC-GPU, was used to project the three-dimensional phantoms into realistic synthetic mammograms. 2000 mammograms with 2522 masses were generated to augment a real data set during training. From the Curated Breast Imaging Subset of the Digital Database for Screening Mammography (CBIS-DDSM) data set, we used 1111 mammograms (1198 masses) for training, 120 mammograms (120 masses) for validation, and 361 mammograms (378 masses) for testing. We used faster R-CNN for our deep learning network with pretraining from ImageNet using the Resnet-101 architecture. We compared the detection performance when the network was trained using different percentages of the real CBIS-DDSM training set (100%, 50%, and 25%), and when these subsets of the training set were augmented with 250, 500, 1000, and 2000 synthetic mammograms. Free-response receiver operating characteristic (FROC) analysis was performed to compare performance with and without the synthetic mammograms. We generally observed an improved test FROC curve when training with the synthetic images compared to training without them, and the amount of improvement depended on the number of real and synthetic images used in training. Our study shows that enlarging the training data with synthetic samples can increase the performance of deep learning systems.

2D/3D reconstruction of the distal femur using statistical shape models addressing personalized surgical instruments in knee arthroplasty: A feasibility analysis.

Personalized surgical instruments (PSI) have gained success in the domain of total knee replacement, demonstrating clinical outcomes similar or even superior to both traditional and navigated surgeries. The key requirement for prototyping PSI is the availability of the digital bony surface. In this paper, we aim at verifying whether the 2D/3D reconstruction of the distal femur, based on statistical shape models (SSM), grants sufficient accuracy, especially in the condylar regions, to support a PSI technique. Computed tomographic knee datasets acquired on 100 patients with severe cartilage damage were retrospectively considered in this work. All the patients underwent total knee replacement using the PSI-based surgical technique. Eighty out of 100 reconstructed distal femur surfaces were used to build the statistical model. The remaining 20 surfaces were used for testing. The 2D/3D reconstruction process was based on digital reconstructed radiographies (DRRs) obtained with a simulated X-ray projection process. An iterative optimization procedure, based on an evolutionary algorithm, systematically morphed the statistical model to decrease the difference between the DRR, obtained by the original CT dataset, and the DRR obtained from the morphed surface. Over the 80 variations, the first ten modes were found sufficient to reconstruct the distal femur surface with accuracy. Using three DRR, the maximum Hausdorff and RMS distance errors were lower than 1.50 and 0.75mm, respectively. As expected, the reconstruction quality improved by increasing the number of DRRs. Statistical difference (P<0.001) was found in the 2 vs 3, 2 vs 4 and 2 vs 5 DRR, thus proving that adding just a single displaced projection to the two traditional sagittal and coronal X-ray images improved significantly the reconstruction quality. The effect of the PSI contact area errors on the distal cut direction featured a maximum median error lower than 2° and 0.5° on the sagittal and frontal plane, respectively. Statistical difference was found (P<0.0001) in the reconstruction accuracy when comparing SSM built using pathologic with respect to non-pathologic shapes (cadavers), meaning that, to improve the patient-specific reconstruction, the morphologic anomalies, specific to the pathology, must be embedded into the SSM. We showed that the X-ray based reconstruction of the distal femur is reasonable also in presence of pathologic bony conditions, featuring accuracy results similar to earlier reports in the literature that reconstructed normal femurs. This finding discloses the chance of applying the proposed methodology to the reconstruction of bony surfaces used in the PSI surgical approach.

Description and characterization of a novel method for partial volume simulation in software breast phantoms.

A modification to our previous simulation of breast anatomy is proposed to improve the quality of simulated x-ray projections images. The image quality is affected by the voxel size of the simulation. Large voxels can cause notable spatial quantization artifacts; small voxels extend the generation time and increase the memory requirements. An improvement in image quality is achievable without reducing voxel size by the simulation of partial volume averaging in which voxels containing more than one simulated tissue type are allowed. The linear x-ray attenuation coefficient of voxels is, thus, the sum of the linear attenuation coefficients weighted by the voxel subvolume occupied by each tissue type. A local planar approximation of the boundary surface is employed. In the two-material case, the partial volume in each voxel is computed by decomposition into up to four simple geometric shapes. In the three-material case, by application of the Gauss-Ostrogradsky theorem, the 3D partial volume problem is converted into one of a few simpler 2D surface area problems. We illustrate the benefits of the proposed methodology on simulated x-ray projections. An efficient encoding scheme is proposed for the type and proportion of simulated tissues in each voxel. Monte Carlo simulation was used to evaluate the quantitative error of our approximation algorithms.

Development and characterization of an anthropomorphic breast software phantom based upon region-growing algorithm

We present a novel algorithm for computer simulation of breast anatomy for generation of anthropomorphic software breast phantoms. A realistic breast simulation is necessary for preclinical validation of volumetric imaging modalities. The anthropomorphic software breast phantom simulates the skin, regions of adipose and fibroglandular tissue, and the matrix of Cooper's ligaments and adipose compartments. The adipose compartments are simulated using a seeded region-growing algorithm; compartments are grown from a set of seed points with specific orientation and growing speed. The resulting adipose compartments vary in shape and size similar to real breasts; the adipose region has a compact coverage by adipose compartments of various sizes, while the fibroglandular region has fewer, more widely separated adipose compartments. Simulation parameters can be selected to cover the breadth of variations in breast anatomy observed clinically. When simulating breasts of the same glandularity with different numbers of adipose compartments, the average compartment volume was proportional to the phantom size and inversely proportional to the number of simulated compartments. The use of the software phantom in clinical image simulation is illustrated by synthetic digital breast tomosynthesis images of the phantom. The proposed phantom design was capable of simulating breasts of different size, glandularity, and adipose compartment distribution. The region-growing approach allowed us to simulate adipose compartments with various size and shape. Qualitatively, simulated x-ray projections of the phantoms, generated using the proposed algorithm, have a more realistic appearance compared to previous versions of the phantom. A new algorithm for computer simulation of breast anatomy has been proposed that improved the realism of the anthropomorphic software breast phantom.

Three-Dimensional Fracture Simulation for Preoperative Planning and Education

Background: Mental translation of the fracture pattern seen on two-dimensional (2-D) radiographs into the real three-dimensional (3-D), intraoperative situation may be difficult. In this study, a concept and a software for 3-D fracture simulation on a personal computer (PC) are presented enabling the surgeon to draw the fracture shown on plain radiographs onto a 3-D model of similar bone and to do virtual reduction and fixation. Material and Methods: A 2-D/3-D database of helical CT volume data from cadaveric tibiae with the corresponding 2-D radiographs was built. A new procedure was used to select bone being most similar to the patient's bone from the database. Tools for drawing the fracture on the 3-D bone model and for individual manipulation of 3-D objects – fragments and/or implants – were developed. Results: After having found the most similar bone in the database, a three-stage fracture simulation can be performed: fracture generation, fragment separation, and fragment manipulation. Each piece of the fractured bone can be individually translated or rotated around a specific point or axis in 3-D space. The corresponding simulated X-ray projection images can be generated from any perspective according to the fragment position and compared with the original radiographs. Simulation of internal fixation is also possible, including implant models. Conclusions: 3-D representation of a fracture may be helpful in understanding fracture morphology as well as reduction mechanisms even in cases where 3-D imaging would normally not be performed. The presented concept can be used for preoperative planning and for educational purposes. The presented software runs on a low-cost PC and may therefore support the surgeon's daily work in the near future.